Voter Registration Costs and Disenfranchisement:
Experimental Evidence from France
Online Appendix
Céline Braconnier
∗
Jean-Yves Dormagen
†
Vincent Pons
‡
February 2016
Appendix A: Sampling frame
In each precinct, addresses and apartments in which unregistered and misregistered citizens were likely to reside were identied as follows.
We rst collected the list of citizens
registered at the precinct as of January 2011 and ordered it by address. Between May and
September 2011, surveyors went to each address and wrote down names found on the mailboxes or on intercoms and the corresponding apartment numbers. This preliminary work was
conducted at 6,030 addresses, excluding addresses that were not found or were inaccessible
to the canvassers. When all names found on a mailbox also appeared on the voter roll, we
excluded the corresponding apartment from the experiment, given the low probability of nding unregistered or misregistered citizens there. In 17 percent of addresses, it was impossible
to link apartments to mailboxes, due to the lack of any number or available identication,
so that all apartments were covered by canvassers, whether included in the sample or not.
Overall, 20,502 apartments likely to host unregistered or misregistered citizens, located at
4,118 addresses, were included in the experimental sample.
Université de Cergy-Pontoise
Université Montpellier 1
‡
Harvard Business School
∗
†
1
Appendix B: Estimate of the initial numbers of unregistered and
misregistered citizens
Studies of voter turnout can use the voter rolls as their sample. Unfortunately, these rolls
are of little help when it comes to studying unregistered and misregistered citizens. Indeed,
there is no systematic list of all citizens at each address who are eligible to register, to
which the voter rolls could be compared. We can nonetheless estimate the initial numbers of
unregistered and misregistered citizens based on the information collected by the canvassers
during the door-to-door visits.
For each apartment that opened its door, the canvassers
reported the number of well-registered, misregistered and unregistered citizens, as well as the
number of foreigners.
We rst focus on addresses for which it was possible to link apartments to mailboxes. In
this group of addresses, we were able to collect monitoring sheets for addresses accounting
for 89 percent of all apartments. The canvassers were able to identify the number of wellregistered citizens, misregistered citizens, unregistered citizens, and foreigners in 67 percent
of these apartments.
We address three issues when examining this data.
First, the apartments that opened
their door might systematically dier from those that remained closed. Second, some citizens
who were not on the 2011 voter rolls had already registered by the time of the visits. Third,
the information recorded by the canvassers might be subject to systematic bias.
1st caveat : Non-representativeness of apartments which opened their door
Apartments with relatively more individuals may be more likely to open their door. Failing to take this into account would result in overestimating the size of the average household.
More generally, any systematic dierence between apartments that opened their door and
the others might introduce biases when estimating the number of citizens of each type. We
address this diculty by comparing the household composition of apartments which opened
2
their door twice vs. only once in addresses that were targeted for two visits.
The strategy is as follows. Whether or not the apartment hosts at least one unregistered
citizen can be described by a dummy. Similarly, whether it hosts exactly one, two, three or
more unregistered citizens can be described by a series of dummies.
The same is true, of
course, for well-registered citizens, misregistered citizens, and foreigners.
For any dummy, denote by
Ni
and
fi
ments for which the dummy is equal to
the (unobserved) number and proportion of apart-
i = (0
or
1).
(unobserved) probabilities that an apartment of type
Denote
i
p1i , p2i , p1&2
i
and
ρi
to be the
opens its door in phase 1 (early vis-
its), in phase 2 (late visits), or in both phases and the correlation between the two Bernoulli
random variables describing door-opening in phase 1 and 2. We assume that
that
ρ1 = ρ0 = ρ:
p1i = p2i
and
the correlation between the two Bernoulli variables is the same for apart-
ments of type 0 and of type 1. Denote
N , Xi1
and
Xi1&2
of apartments and the numbers of apartments of type
i
to be the (observed) total number
which opened their door in phase 1
or in both phases.
This gives us a system of 4 equations and 5 unknowns:




X01 = N0 p10







X11 = (N − N0 ) p11



X01&2 = N0 p1&2

0






X11&2 = (N − N0 ) p1&2
1
But
(1)
(2)
(3)
(4)
1
p1&2
can be written as a function of pi and ρ for i = (0
i
or
2
1): p1&2
= (p1i ) +ρ×p1i (1 − p1i ).
i
This reduces the number of unknowns to 4 and makes the system solvable.
We nd that
N0
must be a solution to the following equation:
2 1 1&2
1 1&2
(N
)
X
X
−
X
X
=0
0
1
0
0
1
h
i
1 1&2
1 1&2
1 2
1
1&2
1 2
1
1&2
+ N0 N X0 X1 − X1 X0
− (X0 ) X1 − X1
− (X1 ) X0 − X0
h
i
2
+
N (X01 ) X11 − X11&2
This admits two possible roots. N0 is equal to the root between 0 and N .
3
From
N0
we immediately infer
N1 ≡ N − N0
and
f1 ≡
N1 1
N1 +N0
. We repeat this exercise for any dummy, and infer that, according to the information recorded
by the canvassers, the average sample apartment included 0.58 well-registered citizens, 0.30
misregistered citizens, 0.27 unregistered citizens and 0.48 foreigners, for a total of 1.62 adult
members.
2nd caveat: Taking into account the timing of the visits
Some of the citizens identied as well-registered by the canvassers had registered between
1 January 2011 and the visits. Since we know the date of the visits as well as the registration
date, we can precisely estimate the importance of this phenomenon. We nd that, on average,
0.03 citizens who were initially misregistered and 0.02 citizens who were initially unregistered
had registered by the time of the visits.
Combined with the information recorded by the canvassers, this means that, as of January 2011, apartments in the experimental sample included 0.53 well-registered citizens, 0.33
misregistered citizens, 0.29 unregistered citizens and 0.48 foreigners.
3rd caveat: Systematic biases in the information recorded by canvassers
The information recorded by canvassers might be systematically biased. Many citizens are
confused about their registration status. In particular, many do not know that registration
is address-specic and thus claim that they are well-registered even though they moved away
without updating their registration status.
The context of the visits did not enable the
canvassers to ask more questions than what was natural to advise the respondents on what
course of action to take. Moreover, the reemphasized civic norm of participation might have
resulted in some respondents consciously lying about their registration status, to avoid being
perceived as deviants.
1f
can be estimated using this method only if X01&2 6= 0, X11&2 6= 0 and X11 6= 0 or X01 6= 0 . We use the
following approximations for characteristics for which these conditions are not satised: f1 ∼ 1 if X01&2 = 0;
f1 ∼ 0 if X11&2 = 0; f1 ∼ 0 if X01&2 6= 0 and X11&2 6= 0 but X11 = 0; f1 ∼ 1 if X01&2 6= 0 and X11&2 6= 0 but
X01 = 0.
1
4
Altogether, we expect the misidentication of misregistered citizens as registered citizens
to be by far the largest bias. Fortunately, there is a more reliable method to estimate the
number of well-registered citizens. We count the number of citizens listed on the 2011 voter
rolls and whose name was found on the mailbox of an apartment in the experimental sample
and we add the number of citizens who were automatically registered as they turned 18.
2
On
average, we nd that experimental sample apartments host 0.23 well-registered citizens. We
infer that the extra 0.30 (0.53 - 0.23) are actually misregistered citizens.
In the end, our preferred estimate is that initially, experimental sample apartments included
0.23 well-registered citizens, 0.63 misregistered citizens, 0.29 unregistered citizens and 0.48
foreigners.
The composition of the total sample population
While the estimates above were derived for apartments located in addresses in which it
was possible to link apartments to mailboxes, we assume that the number of initially misregistered citizens and unregistered citizens was similar in apartments located in the other
addresses.
Adding the citizens who were automatically registered as they turned 18, we
estimate that apartments in the experimental sample hosted 12,878 misregistered citizens
and 5,977 unregistered citizens.
In addition, 4,422, 15,371 and 4,434 well-registered citi-
zens were hosted respectively by the experimental sample apartments, by non-experimental
sample apartments located in the experimental sample addresses, and by addresses located
in the initial sample but not included in the experimental sample because no additional
name was found on the mailboxes. In sum, in the areas covered by this study, we estimate
that there were initially 24,227 (56.2 percent) well-registered citizens, 12,878 (29.9 percent)
misregistered citizens and 5,977 (13.9 percent) unregistered citizens.
2 This
sum might slightly dier from the actual number of well-registered citizens for two dierent reasons,
with opposite sign. First, this method identies as well-registered citizens people who have actually moved
out but whose name is still on the apartment's mailbox, because a relative still lives there, for instance.
Second, this method fails to identify as well-registered citizens people who live in the apartment, but whose
name is not listed on the mailbox.
5
Fraction of citizens initially unregistered and misregistered who live in apartments that opened their door to canvassers at least once
Based on canvassers' reports, the average apartment targeted for two visits which opened
its door at least once hosts 0.37 well-registered citizens, 0.19 misregistered citizens, 0.17 unregistered citizens and 0.33 foreigners. After applying the same transformations as above, we
estimate that 69.9 percent of citizens initially unregistered or misregistered live in apartments
which opened their door to canvassers.
Appendix C: Formal model
The following model extends the standard cost-benet model of the voting decision
(Downs 1957; Riker and Ordeshook 1968) to account for registration as a rst separate
stage and model its connection with the second stage, voting. In addition, we describe likely
type dierences between compliers (citizens registered as a result of the visits) and alwaystakers (newly registered citizens who would have registered regardless of whether or not they
receive a visit) along two dimensions that explain individuals' decisions to register and vote:
benets of voting and the registration cost.
Two stages: registration, and voting
Each unregistered citizen needs to decide whether to register and second, whether to vote.
Individual
of voting
i
bi . c i
is characterized by her net registration cost
ci
and her average net benets
includes gathering information about the registration process and actually
going through the process. It is higher for those who are less comfortable with bureaucratic
tasks, who live further away from the town hall or work during opening hours, who have
unconventional living situations that do not easily meet residency requirements, or who
move frequently and thus have to re-register more often.
ci
may also depend on the person's
wealth: a given time spent to go to the town hall and register imposes a higher monetary
6
cost on the rich, but it may impose a higher utility cost on the poor, whose marginal utility
of consumption is higher (e.g., Alatas et al. 2013).
bi
includes expressive and instrumental
benets, minus the cost of voting. For simplicity, we assume that there is only one electoral
round and that there is no intertemporal actualization rate.
In the rst stage, if
i
registers, she has to pay
g (bi ). i decides to register if ci ≤ g (bi ).
λci
cost decreases to
with
In the second stage,
vote if
Fε ,
bi + εi ≥ 0,
and
i
where
E [εi ] = 0. λ ∈ [0, 1),
ci
and expects to get second-stage utility
If she receives the visit of canvassers, her registration
and
i
decides to register if
λci ≤ g (bi ).
can cast a vote if she registered in the rst stage. She decides to
εi
is a shock realized after registering, with density
fε ,
distribution
represents all factors that aect the benets of voting and which
are unknown at the time of registering, including, for instance, corruption scandal aecting
the candidate
i
was planning to vote for; new polls aecting her expectations about the
closeness of the election; transition to or from unemployment which aects her views about
the general economic situation; unexpected travel plans which force her to be absent on
election day thereby increasing the cost of voting.
We infer that
i's
second-stage utility, conditional on being registered, is
ˆ
∞
g (bi ) ≡
(bi + ε) fε (ε)dε.
−bi
Her propensity to vote, conditional on being registered, is
v(bi ) ≡ P (bi + εi ≥ 0) = 1 − Fε (−bi )
such that
v (b)
and
g (b)
both increase in
b.
Two simple cases: uniform benets of voting or registration cost
Let us now analyze the dierences between compliers and always-takers along benets
of voting and the registration cost.
Since the compliers only register when registration is
7
facilitated, we expect them to be characterized by lower benets of voting and/or a higher
registration cost on average. This is indeed the conclusion that we reach when we consider two
simple cases, where benets of voting or registration cost are uniform across all individuals.
i's (bi = b).
We rst consider the case where the benets of voting are uniform across all
Always-takers and compliers are characterized respectively by
ci ≤ g (b)
g (b) < ci ≤
and by
g(b)/λ (see Figure A4a). Compliers face a higher registration cost than always-takers, but have
identical benets of voting and the same propensity to vote, conditional on being registered.
i's (ci = c).
We next consider the case where the registration cost is uniform across all
The always-takers are then characterized by
bi < g −1 (c)
g −1 (c) ≤ bi
and the compliers by
g −1 (λc) ≤
(Figure A4b). The visits result in the registration of citizens who face the same
registration cost as always-takers but have lower benets of voting and a lower propensity to
vote, conditional on being registered.
General case
In the more general case, in which both benets of voting and registration cost vary across
citizens, it is not necessarily the case that the compliers are characterized by lower benets
of voting and/or a higher registration cost than always-takers.
The distribution of types over the entire population of unregistered citizens is now described by the continuous bivariate random vector of benets of voting and registration costs
(B, C),
with joint density function
f (b, c)
The always-takers are characterized by
and marginal density functions
fB (b)
and
fC (c).
ci ≤ g (bi ) and the compliers by g (bi ) < ci ≤ g(bi )/λ
(Figure A4c). Among citizens facing a given registration cost, it is immediate that compliers
have lower expected benets of voting than always-takers. Similarly, among citizens with a
given expected benet of voting, compliers face a higher registration cost than always-takers.
However, these results do not mechanically extend to the comparison of all compliers and
always-takers.
function
f (b, c)
As an example, consider the case represented in Figure A4d.
is such that
g(bi ) ≤ g1
or
g(bi ) ≥ g2
any
i.
In addition, for all
g(bi ) ≤ g1 , ci ≤ g(bi ); and for all i such that g(bi ) ≥ g2 , ci ≥ g(bi ).
8
The density
i
such that
Then, all the always-takers
have benets of voting lower than
than
g2 :
g1 ,
and all the compliers have benets of voting higher
on average, compliers have higher benets of voting than always-takers. It is equally
easy to construct density functions such that, on average, compliers have a lower registration
cost than always-takers.
It is, however, possible to identify a set of sucient conditions that rule out these cases,
including the condition that
−f (b, c)
satises log-increasing dierences in
b
and
c.
Under
these conditions, we obtain that compliers have lower benets of voting and face a higher
registration cost on average than always-takers. They have a lower propensity to vote, and
those who vote have lower benets of voting. The full statement of all conditions and the
formal derivation of these results are available upon request.
Appendix D: Controlling for compositional eects in turnout estimations
Dierences between the propensity to vote of compliers and always-takers might capture
compositional eects. Indeed, as shown in Section 3.2, the impact of the visits was larger
among unregistered than misregistered citizens. As a result, the compliers account for relatively more citizens who were initially unregistered than the always-takers.
But citizens
with dierent initial registration statuses might have dierent propensities to vote.
To
compare compliers and always-takers who share the same initial registration status, we
allow the
α+
P4
r=1
γ
and the
r
γ r Ni,b
+
mies equal to 1 if
i
δt 's
P6
to vary by initial registration status
r t
t=1 δt Tb
r
× Ni,b
+i,b
r
in Equation [2]:
Vi,b =
1
2
3
4
(3), where Ni,b
, Ni,b , Ni,b and Ni,b are dum-
is newly registered and if she was, respectively, previously unregistered,
registered in another city, registered at another address in the same city, or automatically
registered.
The results are presented in Table A3.
On average, controlling for the initial
registration status, the propensity to vote of newly registered citizens was 2.7 points lower in
the treatment groups than in the control group. The dierence with the estimate we obtain
without controlling for initial registration status (2.2) is not statistically signicant (p-value
9
of 0.88).
Similarly, in Table A5, we control for the initial registration status when comparing
the percent turnout decline between the presidential and the general elections between newly
registered citizens in the control and treatment groups. We rst compute the percent turnout
decline among newly registered citizens by treatment group and initial registration status
(previously unregistered, registered in another city, registered at another address in the same
city, or automatically registered). We then compute the weighted average of the dierence
in turnout decline between the control and treatment groups across newly registered citizens
with dierent initial registration statuses. Each weight corresponds to the fraction of citizens
with a particular initial registration status. We nd that the turnout decline was larger by
3 percentage points among newly registered citizens in the treatment groups, a dierence
signicant at the 10 percent level.
Appendix E: Sampling frame of the postelectoral survey
The postelectoral survey was administered between June 18, the day following the second
round of the general elections, and July 15. All 50 surveyors were students in political science,
economics, social sciences, or law. To facilitate the coordination of the surveyors, the survey
took place in only four cities, Saint-Denis, Cergy, Sevran and Montpellier, which account for
84 percent of the entire sample.
The survey was administered only to French citizens who were not registered at their address
as of January 2011. For this purpose, for each address, surveyors were given a list of names
of individuals that they should NOT survey: citizens who were registered on the 2011 voter
rolls and citizens who were automatically registered in 2011. After introducing themselves
and explaining the purpose of their visit, the surveyors asked the person who had opened
the door whether he was a French citizen. If yes, they asked him whether he accepted to
10
respond, wrote down his rst and last name and rapidly checked that he was not listed on
their list. If not, they went on administering the questionnaire. If their interlocutor was not
French, not willing to answer, or if his name appeared on the list, they asked whether they
could survey another member of the household. Surveyors were instructed to survey no more
than one person in each apartment. In total, the survey plan included 14030 apartments and
1465 surveys were conducted. The fraction of apartments in which one survey was conducted
was not signicantly dierent between the control and treatment groups.
The surveyors did not know the treatment condition of the buildings where they conducted
surveys. Still, we could not exclude ex ante that the response rate might be dierent in the
control and treatment groups. Therefore, half of the addresses were randomly selected to be
covered twice: in these addresses, surveyors knocked again at all doors that had remained
closed the rst time. Since we do not nd any statistically signicant dierence between the
answer rates in the control and in the treatment groups, we do not exploit this feature when
analyzing the data.
Finally, administering the questionnaire required 15 to 20 minutes on average. Only 2 percent
of respondents who started answering the questionnaire refused to go to the end.
Appendix F: Prediction of political preferences based on demographics
To predict dierences between the political preferences of the newly registered and the previously registered citizens and between the compliers and always-takers based on their demographics, we proceed in three steps.
First, we regress the preferences expressed by the
respondents to the postelectoral survey on three demographic characteristics available on the
voter rolls for all registered citizens, as specied in the following equation:
Lefti,b
= α1 + α2 Genderi,b + α3 Agei + α4 Immigranti,b + i
11
where Lefti,b is a dummy equal to 1 if the respondent located himself on the left of the
left-right axis or had a preference for a left candidate (and 0 if he located himself on the
right), Genderi,b is equal to 1 if the respondent is a male and Immigranti,b is equal to 1 if the
respondent is an immigrant. The results are presented in Table A10, Panel A. Age and being
an immigrant are strong predictors of preference on the left, and have the expected sign.
Second, we use the estimated coecients
α
c1 , α
c2 , α
c3
and
α
c4
to predict the political pref-
di,b .
erences of all registered citizens in the sample, Left
Third, we estimate dierences between the predicted political preferences of the newly
registered and the previously registered citizens and between the compliers and always-takers.
Formally, we estimate the following model:
di,b
Left
where
Ni,b
= α + βNi,b + δTb × Ni,b + i,b
is a dummy equal to 1 if
i
is a newly registered citizen.
Table A10, Panel B
performs this analysis.
Appendix G: Estimate of the eect of the intervention Early Home
registration & Late Home registration on overall turnout
To estimate the eect of the intervention Early Home registration & Late Home registration on overall turnout, we rst compute the fraction of citizens initially unregistered and
misregistered who live in apartments that opened their door to canvassers at least once: 69.9
percent (see Appendix 3). The eect of the intervention on citizens who live in apartments
which actually opened their door should thus be scaled by 1/0.699
= 1.43.
We infer from the
estimates presented in Table 1 and Table 2 that the intervention increased electoral participation from 16.1 percent to 29.2 percent at the rst round of the presidential election. Indeed,
12
the average apartment hosts 0.92 initially unregistered and misregistered citizens.
In the
control group, the number of new registrations in the average apartment was 0.168 (Table
1), and 87.5 percent of the newly registered citizens participated in the rst round of the
presidential elections (Table 2). Thus, in the control group, the participation of all citizens
initially unregistered and misregistered was
0.168
0.92
× 87.5% = 16.1%.
Instead, in the group
Early Home registration & Late Home registration, the visits increased the number of new
registrations by 0.096 in the average apartment, and by
0.096 × 1.43 = 0.137
in apartments
which opened their door at least once. Of the newly registered citizens in that group, 87.3
percent participated in the rst round of the presidential elections. Thus, the participation
of all citizens initially unregistered and misregistered and living in these apartments was
0.168+0.137
0.92
× 87.3% = 29.2%.
Similarly, the intervention increased electoral participation from
16.4 percent to 29.3 percent at the second round of the presidential elections, and from 9.8
percent to 14.5 percent and 8.9 percent to 15.6 percent at the general elections among these
citizens. We assume that the eect would be the same among citizens living in apartments
which did not open their door at either the rst or second visit.
The above gures represent increased participation at the polling station closest to each
citizen's place of residence. However, a fraction of the citizens who remained misregistered
at the end of the registration period participated in the elections by travelling back to their
previous address or by voting by proxy.
We do not observe their participation rate, but
can estimate it based on the observed participation of their counterparts: citizens who are
registered here but live elsewhere (as signaled by the fact that their name was not found on any
mailbox). The implicit assumptions here are that the participation of misregistered citizens
who move out is similar to the participation of those who move in, and that the participation
rate of misregistered compliers would have been identical to the participation rate of other
misregistered citizens had they not registered. The latter assumption is valid to the extent
that the decision to register, by misregistered compliers, signals a higher cost of voting at the
previous address (predicting lower participation) as much as a higher interest in the elections
13
(predicting higher participation). The participation of citizens who are registered here but live
elsewhere was 58.4 percent and 61.3 percent at the presidential elections, and 37.2 percent and
35.9 percent at the general elections. The fraction of citizens who were initially unregistered
or misregistered and are still misregistered is 57 percent in the control group and 50 percent in
the group Early Home registration & Late Home registration. We infer that this intervention
increased electoral participation among citizens initially unregistered or misregistered from
49.4 percent to 58.4 percent and 51.3 percent to 60 percent at the presidential elections, and
from 31 percent to 33 percent and 29.4 percent to 33.5 percent at the general elections.
As a nal step, we have to factor in the participation of well-registered citizens: 76.6
percent and 78.5 percent at the presidential elections, and 49.1 percent and 47.2 percent at
the general elections. Taking into account the relative shares of the dierent categories of
citizens in the areas covered by this study, we conclude that the Early Home registration
& Late Home registration increased overall participation from 64.7 percent to 68.6 percent
and 65.6 percent to 69.3 percent at the presidential elections, and from 41.2 percent to 42.1
percent and 39.4 percent to 41.2 percent at the general elections.
14
Appendix H: Appendix tables and gures
Figure A1. Turnout at French Presidential and General elections, 1988-2012
Figure A2: Localization of the 10 cities included in the experiment
Sevran
Cergy St-Denis
Paris
Le Taillan
Blanquefort
Lormont
Eysines
Montpellier
Carcassonne
15
Figure A3: Example of leaets handed out by the canvassers
16
Figure A4. Graphic representation of the different cases discussed in the model
Always‐takers
Compliers
A3a. Uniform benefits of voting
A3b. Uniform registration costs
g(bi)
g(bi)
Always‐takers
Always‐takers
Compliers
Compliers
g(b)
c
λc
ci
g(b)
c
g(b)/λ
A3c. General case
ci
A3d. Compliers with higher benefits of voting than
always‐takers
Always‐takers
g(bi)
g(b)=c
Compliers
g(bi)
g(b)=c
Always‐takers
Compliers
g(b)= λc
g(b)= λc
ci
ci
Table A1: Summary statistics
Any treatment
Control group
Panel A. At the address level
Number of mailboxes
Number of apartments included in sample Housing price
Panel B. At the apartment level
Number of additional names on mailbox Panel C. At the individual level
Age
Gender
In couple
Number of other household members
Education
No diploma
Less than end‐of‐high‐school End‐of‐high‐school More than end‐of‐high‐school Activity
Employed
Unemployed
Inactive
Housing situation
Owner
Tenant, social housing Tenant, private housing Personal monthly income Less than 700 euros 700 ‐ 1100 euros 1100 ‐ 1500 euros Above 1500 euros Born in France
Born in same département
Was naturalized French
Holds another citizenship
Speaks French with family members French only
Some French, some other language
Other language only
Has lived in the city
For 2 years
2 ‐ 5 years
5 ‐ 10 years
More than 10 years
Adherent of a religion
Regular churchgoer
Treatment groups
Treatment groups included separately
Test: joint significance of treatment dummies
P‐value
Number of obs. Mean
SD
Mean SD
P‐value Treatment = Control
7.9
5.1
3103
11.0
7.7
871
7.8
4.9
3150
10.3
7.0
874
0.661
0.600
0.476
0.995
0.993
0.977
4118
4118
941
1.3
0.7
1.3
0.7
0.213
0.747
20502
36.3
0.403
0.543
1.9
13.6
0.491
0.499
1.6
36.3
0.425
0.523
2.0
13.0
0.495
0.500
1.7
0.978
0.461
0.508
0.694
0.016
0.909
0.148
0.357
1450
1464
1458
1463
0.146
0.278
0.256
0.320
0.354
0.449
0.437
0.467
0.146
0.278
0.218
0.357
0.354
0.448
0.413
0.479
0.994
0.994
0.163
0.204
0.831
0.013
0.101
0.222
1450
1450
1450
1450
0.623
0.103
0.274
0.485
0.305
0.447
0.615
0.112
0.273
0.487
0.315
0.446
0.806
0.650
0.970
0.021
0.278
0.364
1458
1458
1458
0.139
0.554
0.307
0.347
0.498
0.462
0.113
0.598
0.289
0.317
0.490
0.453
0.297
0.356
0.681
0.192
0.622
0.148
1440
1440
1440
0.225
0.206
0.260
0.309
0.758
0.246
0.210
0.213
0.418
0.405
0.440
0.463
0.429
0.431
0.408
0.410
0.197
0.210
0.277
0.315
0.753
0.232
0.238
0.234
0.398
0.408
0.448
0.465
0.432
0.422
0.426
0.423
0.312
0.869
0.557
0.840
0.823
0.608
0.295
0.428
0.292
0.823
0.641
0.856
0.468
0.045
0.264
0.114
1281
1281
1281
1281
1455
1450
1393
1404
0.581
0.404
0.014
0.494
0.491
0.118
0.612
0.371
0.017
0.487
0.483
0.130
0.349
0.301
0.671
0.579
0.646
0.696
1457
1457
1457
0.168
0.179
0.156
0.497
0.667
0.355
0.374
0.384
0.364
0.501
0.472
0.479
0.185
0.156
0.157
0.501
0.687
0.323
0.389
0.363
0.364
0.500
0.464
0.468
0.487
0.366
0.970
0.919
0.537
0.330
0.239
0.087
0.310
0.307
0.079
0.770
1458
1458
1458
1458
1414
1373
Notes: For each variable, we report the means and standard deviations in both the control group and in all treatment groups pooled together and indicate the p‐value of the difference. We then take each treatment group separately and test the hypothesis of joint significance of the treatment dummies.
Unit of observation is the address in Panel A, the apartment in Panel B, and the respondent to the post‐electoral survey in Panel C. In Panels B and C, standard errors are adjusted for clustering at the address level.
Table A2: Impact of the interventions on the number of new registrations, by initial registration status
(1)
All newly registered
(2)
Not registered before
(3)
Registered in another city before
0.048***
(0.008)
Yes
Yes
20458
0.03
0.168
0.022***
(0.004)
Yes
Yes
20458
0.02
0.047
0.014**
(0.006)
Yes
Yes
20458
0.04
0.079
0.008***
(0.003)
Yes
Yes
20458
0.02
0.025
0.004*
(0.002)
Yes
Yes
20458
0.02
0.013
Early Canvassing & Late Home
registration
Early Home registration & Late
Home registration
Strata fixed effects
Apartment & Building controls
Observations
R‐squared
Mean in Control Group
0.014
(0.012)
0.031**
(0.012)
0.032**
(0.013)
0.054***
(0.013)
0.060***
(0.013)
0.096***
(0.014)
Yes
Yes
20458
0.03
0.168
0.01
(0.007)
0.006
(0.006)
0.012*
(0.006)
0.022***
(0.007)
0.035***
(0.007)
0.047***
(0.007)
Yes
Yes
20458
0.02
0.047
‐0.005
(0.008)
0.012
(0.009)
0.01
(0.009)
0.020**
(0.008)
0.015*
(0.009)
0.032***
(0.009)
Yes
Yes
20458
0.05
0.079
0.003
(0.004)
0.010*
(0.005)
0.007
(0.005)
0.008*
(0.005)
0.007
(0.004)
0.013**
(0.005)
Yes
Yes
20458
0.02
0.025
0.004
(0.003)
0.004
(0.003)
0.004
(0.003)
0.004
(0.003)
0.005
(0.003)
0.002
(0.003)
Yes
Yes
20458
0.02
0.013
Linear combinations of estimates:
Average effect of Canvassing
1/2 (EC + LC)
0.022**
(0.010)
0.008
(0.005)
0.004
(0.007)
0.006*
(0.004)
0.004*
(0.002)
Average effect of Home registration
1/2 (EH + LH)
0.043***
(0.011)
0.017***
(0.005)
0.015**
(0.007)
0.007*
(0.004)
0.004
(0.002)
Difference between average effect of Home reg. and Can.
1/2 (EH + LH) ‐ 1/2 (EC + LC)
0.021*
(0.011)
0.009
(0.006)
0.012*
(0.007)
0.001
(0.004)
0.000
(0.003)
Difference between average effect of Late visit and Early visit
1/2 (LH + LC) ‐ 1/2 (EH + EC)
0.020*
(0.011)
0.003
(0.006)
0.013*
(0.007)
0.004
(0.004)
0.000
(0.003)
Panel A. All treatments pooled together
Any treatment
Strata fixed effects
Apartment & Building controls
Observations
R‐squared
Mean in Control Group
Panel B. Each treatment included separately Early Canvassing
Late Canvassing
Early Home registration
Late Home registration
(4)
(5)
Registered at "Automatically" another address registered
in this city before
Notes: Unit of observation is the apartment. We include all newly registered citizens in the sample apartments (column 1), those who were not registered before (column 2), those who were registered in another city before (column 3), those who were registered at another address in the same city before (column 4) and those who were "automatically" registered (column 5). Control variables include: number of mailboxes in the building and number of last names found on the mailbox of the apartment that were absent from the 2011 voter rolls. Clustered standard errors in parentheses. ***, **, * indicate significance at 1, 5 and 10%. Table A3: Electoral participation of citizens by registration status, treatment group, and previous registration status
Newly reg., previously not reg. x Any treatment (1)
Newly reg., previously reg. in another city x Any treatment (2) Newly reg., previously reg. at another address in this city x Any treatment (3) Newly reg., automatically reg. x Any treatment (4) Newly reg., previously not reg.
Newly reg., previously reg. in another city Newly reg., previously reg. at another address in this city Newly reg., automatically reg.
Constant
Observations
R‐squared
Linear combinations of estimates:
Av. difference between newly reg. in treatment gr. and control, controlling for previous reg. status (Weighted average of (1), (2), (3) and (4))
(1)
(2)
Presidential elections
1st round
2nd round
0.012
‐0.023
(0.021)
(0.019)
‐0.048***
‐0.040***
(0.013)
(0.015)
‐0.001
‐0.005
(0.030)
(0.025)
0.034
‐0.019
(0.045)
(0.041)
0.171***
0.179***
(0.019)
(0.017)
0.242***
0.209***
(0.011)
(0.012)
0.189***
0.185***
(0.027)
(0.022)
‐0.048
0.017
(0.039)
(0.036)
0.703***
0.725***
(0.003)
(0.003)
33773
33772
0.02
0.02
‐0.011
(0.011)
‐0.027**
(0.011)
(3)
(4)
General elections
1st round
2nd round
‐0.061*
‐0.005
(0.034)
(0.034)
‐0.041
‐0.038
(0.030)
(0.029)
‐0.029
‐0.069
(0.047)
(0.046)
‐0.043
‐0.016
(0.043)
(0.042)
0.049
0.042
(0.031)
(0.030)
0.136***
0.098***
(0.026)
(0.025)
0.172***
0.191***
(0.041)
(0.040)
‐0.116***
‐0.145***
(0.038)
(0.037)
0.447***
0.430***
(0.004)
(0.004)
33788
33754
0.01
0.01
‐0.046**
(0.019)
‐0.028
(0.019)
(5)
(6)
Average on One vote at least
all rounds
‐0.020
0.010
(0.020)
(0.015)
‐0.040**
‐0.007
(0.016)
(0.009)
‐0.023
‐0.001
(0.027)
(0.019)
‐0.010
‐0.036
(0.030)
(0.036)
0.111***
0.154***
(0.018)
(0.015)
0.171***
0.186***
(0.014)
(0.008)
0.181***
0.165***
(0.024)
(0.017)
‐0.073***
0.047
(0.026)
(0.031)
0.577***
0.786***
(0.003)
(0.003)
33665
33665
0.02
0.02
‐0.027**
(0.011)
‐0.004
(0.008)
Notes: Unit observation is the individual participation at a given electoral round. We include all previously registered citizens (registered before 2011) and all newly registered citizens (registered in 2011). Previously registered citizens are the omitted category. Newly registered citizens are included separately, according to their former registration status. Sample size is slightly smaller than in Table 2 since we drop a few newly registered citizens whose previous registration status is unknown.
We estimate differences in the propensity to vote of newly registered citizens in the control and the treatment groups. Column 6: "One vote at least" is equal to 1 if the individual participated in any of the four rounds. Clustered standard errors are in parentheses. ***, **, * indicate significance at 1, 5 and 10%. Table A4: Percent decline in turnout between the presidential and general elections, by registration status and treatment group
Panel A. Comparison between newly registered citizens and previously registered citizens Previously reg. citizens, all groups Newly reg. citizens, control group Difference between newly reg. citizens and previously reg. Citizens (1)
‐0.384
(0.005)***
‐0.428
(0.016)***
‐0.044
(0.016)***
Panel B. Comparison between newly registered citizens in the treatment groups and in the control group (1)
All treatment gr.
Newly reg. citizens, treatment groups ‐0.453
(0.008)***
Difference between newly reg. citizens in treatment groups and control group ‐0.025
(0.018)
‐0.030
Difference between newly reg. citizens in treatment groups and control group, controlling for initial registration status (0.018)*
(2)
Canvassing gr.
‐0.434
(0.016)***
‐0.006
(0.022)
‐0.011
(0.023)
(3)
Home registration gr.
‐0.467
(0.013)***
‐0.039
(0.021)*
‐0.042
(0.021)**
Notes: We report the point estimates and standard errors of non‐linear combinations of coefficients obtained after running seemingly unrelated regressions of Equation [2]. Panel A estimates the turnout decline between the presidential and general elections among previously registered citizens and among newly registered citizens in the control group. As an example of how to read the table, the coefficients in Panel A mean that the participation of previously registered citizens declined by 38.4% between the presidential and general elections. Newly registered citizens in the control group experienced a decline of 42.8%, 4.4 percentage points stronger than the previously registered. Panel B estimates the turnout decline among newly registered citizens in the control group and treatment groups. The last line reports the weighted average of the difference between participation decline for newly registered citizens with different initial registration status in the treatment and control groups.
***, **, * indicate significance at 1, 5 and 10%. Table A5: Percent decline in turnout between the presidential and general elections among newly registered citizens by treatment group and previous
registration status
(1)
All treatment gr.
Panel A. Newly reg. citizens who were previously unregistered Control group Treatment groups Difference between treatment groups and control group Difference between treatment groups and control group ‐0.49
(0.014)***
‐0.034
(0.031)
‐0.527
(0.021)***
‐0.07
(0.035)**
‐0.423
(0.013)***
‐0.015
(0.026)
‐0.409
(0.023)***
‐0.39
(0.025)***
0.019
(0.033)
‐0.434
(0.021)***
‐0.026
(0.031)
‐0.364
(0.021)***
‐0.052
(0.042)
‐0.312
(0.036)***
‐0.346
(0.036)***
‐0.034
(0.051)
‐0.381
(0.038)***
‐0.069
(0.052)
‐0.605
(0.021)***
‐0.046
(0.051)
‐0.559
(0.046)***
‐0.619
(0.033)***
‐0.059
(0.057)
‐0.546
(0.038)***
0.013
(0.059)
Panel C. Newly reg. citizens who were previously registered at another address in this city Control group Treatment groups Difference between treatment groups and control group Panel D. Newly reg. citizens who were automatically registered Control group Treatment groups Difference between treatment groups and control group (3)
Home registration gr.
‐0.456
(0.028)***
‐0.476
(0.029)***
‐0.02
(0.040)
Panel B. Newly reg. citizens who were previously registered in another city Control group Treatment groups (2)
Canvassing gr.
Notes: We report the point estimates and standard errors of non‐linear combinations of coefficients obtained after running seemingly unrelated regressions of Equation [3]. As an example of how to read the table, the coefficients in Table A mean that the participation of newly registered citizens who previously unregistered declined by 45.6% and by 49% between the presidential and general elections in the control group and in the treatment groups, for a difference of 3.4 percentage points between treatment and control.
***, **, * indicate significance at 1, 5 and 10%. Table A6: Impact of the visits on the participation of citizens registered prior to the visits
Any treatment
Strata fixed effects
Individual and Building controls Observations
R‐squared
Mean in Control Group
(1)
(2)
(3)
(4)
(5)
(6)
Presidential elections
General elections
Average on One vote 1st round 2nd round 1st round 2nd round all rounds at least
‐0.013
‐0.005
‐0.006
0.000
‐0.006
‐0.004
‐0.012
‐0.012
‐0.014
‐0.014
‐0.010
‐0.011
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
8367
8367
8401
8394
8349
8349
0.05
0.05
0.10
0.10
0.09
0.04
0.733
0.752
0.472
0.452
0.602
0.808
Notes: Unit observation is the individual participation at a given electoral round. We include all citizens registered prior to the visits. We estimate differences in the electoral participation of these citizens in the control group and all treatment groups pooled together. Column 6: "One vote at least" is equal to 1 if the individual participated in any of the four rounds. Controls include: age, gender, number of previously registered citizens in the apartment, and number of mailboxes in the building. Clustered standard errors in parentheses. ***, **, * indicate significance at 1, 5 and 10%. Table A7: Characteristics of newly registered citizens in apartments which opened their door for a late home registration visit
(1)
Early Home registration + Late
Home registration
Constant
Observations
R‐squared
(2)
(3)
Individual characteristics
Gender
Age
Born abroad
0.010
(0.040)
0.449***
(0.028)
‐0.243
(1.443)
37.438***
(0.926)
0.097
(0.059)
0.304***
(0.037)
460
0.00
460
0.00
459
0.01
(4)
(5)
Apartment and building characteristics
Number of Number of mailboxes
names of citizens not registered
‐0.004
‐2.182
(0.076)
(2.950)
1.330***
19.383***
(0.059)
(1.901)
460
0.00
460
0.01
Notes: The sample includes all newly registered citizens living in apartments which opened their door at the late visit in the treatment groups "Early Canvassing & Late Home registration" and "Early Home registration & Late Home registration". Omitted group is "Early Canvassing & Late Home registration". We consider individual characteristics (columns 1 through 3) as well as the number of names of citizens not registered initially found on the mailbox corresponding to the person's apartment and the total number of mailboxes and baseline registration rate at her address. Clustered standard errors in parentheses. ***, **, * indicate significance at 1, 5 and 10%. Table A8: Impact of the interventions on level of politicization
(1)
(2)
Index of politicization
Panel A. All treatments pooled together Any treatment
0.047*
(0.024)
No
1465
0.00
0.056**
(0.025)
Yes
1219
0.18
Individual controls
Observations
R‐squared
‐0.019
(0.040)
0.073**
(0.034)
0.095**
(0.038)
0.053
(0.035)
0.046
(0.038)
0.031
(0.039)
No
1465
0.01
0.021
(0.040)
0.090***
(0.035)
0.095**
(0.038)
0.036
(0.035)
0.046
(0.036)
0.044
(0.037)
Yes
1219
0.18
Linear combinations of estimates:
Av. difference between newly registered in Canvassing gr. and control 1/2 (EC + LC)
0.027
(0.030)
0.056
(0.031)*
0.074
(0.030)**
0.065
(0.030)**
0.038
(0.031)
0.045
(0.030)
Individual controls
Observations
R‐squared
Panel B. Each treatment included separately Early Canvassing (EC)
Late Canvassing (LC)
Early Home registration (EH)
Late Home registration (LH)
Early Canvassing & Late Home registration (EC&LH) Early Home registration & Late Home registration (EH&LH) Av. difference between newly registered in Home registration gr. and control
1/2 (EH + LH)
Av. difference between newly registered in Two visits gr. and control
1/2 (EC&LH + EH&LH)
Notes: Unit of observation is the respondent to the post‐electoral survey. The outcome is the standardized average of 36 indicators of level of politicization. Control variables include: gender, age, age squared, unemployed, inactive, less than end‐of‐high‐school diploma, end‐of‐high‐school diploma, higher than end‐of‐high‐school diploma, household size, single, speaks some French and some other language, speaks other language only, tenant in social housing, tenant in private housing, less than 1100 euros income, income between 1100 and 1500 euros, income above 1500 euros, has lived in the city for less than 5 years, between 5 and 10 years, more than 10 years, and country of birth. Clustered standard errors in parentheses. ***, **, * indicate significance at 1, 5 and 10%. Table A9: Impact on the selection operated by the registration process
(1)
(2)
(3)
(4)
(5)
Registered in his Registered (in his Registered at his Average turnout One vote at least
address
city
city or elsewhere)
Joint significativity of all selection variables interacted with…
Panel A. Any treatment
Constant
statistic
p‐value
statistic
p‐value
Observations
R‐squared
315.7
0.000***
65.8
0.000***
1012
0.18
52.9
0.003***
41.4
0.049**
1009
0.11
52.1
0.004***
27.6
0.488
1012
0.09
219.0
0.000***
67.3
0.000***
998
0.12
111.6
0.000***
63.1
0.000***
998
0.10
Panel B. Treatment groups included separately statistic
Constant
p‐value
statistic
Door‐to‐door canvassing group
p‐value
statistic
Home registration group
p‐value
statistic
Two visits group
p‐value
statistic
Home registration group ‐ p‐value
Door‐to‐door canvassing group
statistic
Two visits group ‐ p‐value
Home registration group
Observations
R‐squared
315.7
0.000***
45.8
0.018**
40.9
0.055*
70.1
0.000***
16.8
0.953
43.0
0.035**
1012
0.23
52.9
0.003***
34.6
0.183
41.7
0.046**
58.5
0.001***
37.2
0.115
47.3
0.013**
1009
0.17
52.1
0.004***
33.0
0.235
22.2
0.771
33.5
0.218
30.1
0.358
30.0
0.365
1012
0.14
219.0
0.000***
39.7
0.070*
48.6
0.009***
55.0
0.002***
23.1
0.729
25.4
0.609
998
0.16
111.6
0.000***
45.4
0.020**
41.8
0.046**
69.2
0.000***
29.4
0.394
32.2
0.266
998
0.14
Any treatment group
Notes: Unit of observation is the respondent to the post‐electoral survey. We consider five outcomes: registration in the individual's city (column 1); registration in this or another city (column 2); registration at his address (column 3); average participation at the four electoral rounds of 2012 (column 4); and a dummy equal to 1 if the individual participated in any of the four rounds (column 5). The first and third outcomes are administrative data. The second, fourth, and fifth are self‐reported. We regress individual registration or participation on various individual characteristics and their interaction with treatment dummies. In Panel A, we report the joint significativity of all characteristics and of the characteristics interacted with a treatment dummy. In Panel B, we report the joint significativity of all characteristics, of the characteristics interacted with three treatment dummies (Door‐to‐door canvassing, Home registration, and Two visits), and of the difference between characteristics interacted with two different treatment dummies. ***, **, * indicate significance at 1, 5 and 10%. Table A10: Impact on the political preferences selected by the registration process
(1)
Position on the left
(2)
(3)
(4)
(5)
Vote for left candidate
Presidential elections
General elections
1st round
2nd round
1st round
2nd round
Panel A. Determinants of left/right position and vote choice among respondents to the postelectoral survey
Gender
‐0.036
‐0.005
0.013
‐0.030
0.006
(0.043)
(0.040)
(0.034)
(0.048)
(0.045)
Age
‐0.021
‐0.031**
‐0.030**
0.012
‐0.014
(0.015)
(0.016)
(0.015)
(0.018)
(0.021)
Immigrant
0.151***
0.109***
0.084***
0.155***
0.158***
(0.038)
(0.038)
(0.032)
(0.042)
(0.041)
Constant
0.845***
0.893***
0.951***
0.747***
0.864***
(0.059)
(0.060)
(0.054)
(0.075)
(0.085)
Observations
424
421
415
249
197
R‐squared
0.03
0.02
0.02
0.04
0.05
Panel B. Predicted position on the left and vote shares for the entire sample of registered citizens
Newly registered x Any treatment
0.001
0.001
0.001
‐0.002
(0.003)
(0.003)
(0.002)
(0.005)
Newly registered
0.027***
0.034***
0.032***
‐0.005
(0.003)
(0.003)
(0.002)
(0.004)
Constant
0.773***
0.779***
0.847***
0.837***
(0.001)
(0.001)
(0.001)
(0.001)
Observations
28083
20196
20792
12365
R‐squared
0.02
0.05
0.05
0.00
0.001
(0.005)
0.017***
(0.004)
0.846***
(0.001)
9782
0.01
Notes: In Panel A, the unit of analysis is the respondent to the post‐electoral survey and the outcomes are reported left/right position and vote choice at each of the four rounds. Only respondents who are actually registered in their city are included in the sample and only citizens who voted are included in the sample for the regression of the corresponding electoral round. The outcomes are regressed on all variables available both for respondents to the postelectoral survey and for the entire sample: age, gender, immigrant.
Panel B uses the coefficients estimated in Panel A to predict the left/right position and vote choice of each registered citizen in the four cities included in the survey sample and compares the predicted position of different types of citizens. Only citizens who actually voted are included in the sample for the regression of the corresponding electoral round. For the second round of the general elections, we exclude the cities Saint‐Denis and Sevran, in which only one (left‐wing) candidate remained at the second round. Clustered standard errors in parentheses. ***, **, * indicate significance at 1, 5 and 10%.